AI meeting assistant GeekBye markets itself as undetectable, leaves no shared record

GeekBye, a desktop application that generates real-time AI answers during video calls, markets itself directly on its ability to avoid detection by the other party in the call, according to the company's website.

"They'll never know you're using it. Hidden from screen recordings. No bots join calls. Zero traces," the company states. Elsewhere the product is described as "Completely invisible and undetectable."

The application runs natively on macOS 14 or later and on Windows, and operates inside Google Meet, Microsoft Teams, Zoom, Slack and Webex, according to the site. A user presses Ctrl+Enter, and the software reads the screen and both sides of the audio to generate a response; the company says "responses start streaming within 2-3 seconds" of the keystroke. Suggested replies appear as clickable text rather than executing automatically, and the tool requires access to the user's microphone and the meeting platform's system audio, per the site.

GeekBye markets to "job seekers in technical interviews, sales pros handling objections, students in oral exams, or executives in important meetings," the company states. It also scores a user's performance across six communication metrics with feedback tied to the call transcript, according to the site.

On data handling, GeekBye says transcripts and recordings are stored locally in an encrypted database and are "not uploaded to cloud servers." Pricing runs to $149.99 a month for its Pro tier, with a one-time lifetime-purchase option offered as an alternative. The company states it had "600+ users" as of Aug. 25.

The product bears directly on one dimension of this outlet's trust-gap rubric: audit trail. Where that dimension scores highest for a tamper-evident, customer-reviewable record, GeekBye's own marketing states the product is built to leave no record accessible to the other participant in the conversation. No independent verification of the "undetectable" claim, the stated user count or the response-time figures was found on the page reviewed.